China tops 700 registered generative AI models under CAC filing regime

China’s disclosure that more than 700 generative AI large-model products have completed CAC filing is a strong signal about the depth of its frontier-model ecosystem. Filings aren’t the same as widely-used products, but they track who is serious enough to clear regulatory and infrastructure hurdles. Combined with over a billion HarmonyOS devices, this points towards a tightly integrated hardware–software–model stack that Beijing is treating as strategic infrastructure rather than just a consumer novelty.

For the AGI race, the key point is scale and diversity. Hundreds of filed models imply a long tail of domain-specific and regional systems built atop a handful of major national models. That accelerates diffusion of advanced capabilities into industry and government, while giving regulators leverage via the filing process to shape safety and content norms. It also makes effective decoupling harder: even if Western firms can’t sell leading-edge chips into China, a domestic compute and model ecosystem is clearly taking root. More independent centers of model training mean more parallel experimentation, which tends to shorten timelines to novel capabilities, even if individual models lag the global state of the art.